Te (Thomas) Tang
Imitation LearningTeach robots manipulation skills from human demonstration, then intelligently transfer to different scenarios. In this example, we are teaching robots to manipulate deformable objects (rope or clothes) which requires the adaptability to cope with varying shapes in each trail. Motion adaptation to manipulate flexible objectsMotion PlanningMotion planning on high degree of freedom (DoF) is a challenging task. We need to control the robot arms to do dexterous tasks with real-time vision feedback, while at the same time, avoiding collision with environments or self-collision. Here we demonstrated the motion planning capability on a 18-DOF humanoid system (14-DoF arms, 2-Dof grippers, 1-DoF trunk, 1-DoF head). Dual arm motion planning for wire harness assemblyPerceptionReal-time tracking of featureless deformable objects (no markers). Tracking result was then used to plan the dual-arm robot motion. Teach Robots Assembly from Human DemonstrationFor robotic assembly tasks, tuning the parameters for force controller is a non-trival and time-consuming task. We proposed a novel learning-based framework to teach robots the optimal gains from human demonstration. An Exoskeleton System for Hand RehabilitationHand injuries seriously affect patients’ living qualities. To accelerate the recovery process, we designed a portable exoskeleton system to conduct hand rehabilitation exercise automatically. To make the system extremely light, we innovated a compact motor system based on Ti-Ni shape memory alloy (shrink when heating). |